Nick Hilliard wrote: Someone else mentioned that we are now scraping the bottom of the ipv4 barrel. As of two days ago, there were quantifiable problems associated with 13 out of the 26 remaining /8s. 12 of these are known to be used to one extent or another on internet connected networks, and are seen as source addresses on various end-points around the place. One of them (223/8) has rfc-3330 issues (although later fixed in 5735). So, the issue for IANA is how to allocate these /8s in a way which is demonstrably unbiased towards any particular RIR. The solution which they've agreed on with the RIRs looks unbiassed, unpredictable in advance, calculable in retrospect and best of all, it's not open to abuse. And while Chuck Norris could probably predict the footsie, the djia and the hang-seng weeks in advance, this sort of prognostication appears to be beyond the capabilities of ICANN, IANA and the RIRs. At least if it isn't, no-one's saying anything. Do you have a better suggestion about how to allocate tainted address space in a way that is going to ensure that the organisations at the receiving end aren't going to accuse you of bias? My response: The granularity of allocations is arbitrary, and when scraping the bottom of the barrel, where there are known problems, it may time to get more granular. There's really no difference in managing a handful of /N's rather than /8's, if N is not that much bigger than 8. The granularity boundary probably should stay around or above the biggest assignments a given RIR is expected to make, or has made. So, if the tainted *portions* of problem /8's are set aside, you end up with sets of varying sizes of /N. E.g. if there is one /24 that is a problem, you set that aside, and end up with a set that consists of one each of /9 through /24. Even if you set aside a /16, you get a set which is /9, /10, /11, /12, /13, /14, /15, and /16.
From a documentation and allocation perspective, you could even treat that as giving the whole of the /8 to the RIR, and having them give back (assign) the problem chunk to IANA.
Do this for the 13 problem /8's, and then group the resulting untainted sets and give them out. Give them out according to the needs of the RIRs, and the larger community will consider it fair. No one will think badly of giving the /9's to one of the big 3 (APNIC, ARIN, or RIPE), and the smaller ones to the other RIRs, I'm sure. As long as there are no tainted portions assigned to the RIRs, I don't see how this could be a problem. Brian